Note: Descriptions are shown in the official language in which they were submitted.
CA 02600535 2007-09-28
SYSTEMS AND METHODS FOR ANALYZING COMMUNICATION
SESSIONS USING FRAGMENTS
BACKGROUND
[0001] It is desirable in many situations to record communications, such as
telephone
calls. This is particularly so in a contact center in which many agents may be
handling
hundreds of telephone calls each every day. Recording of these telephone calls
can
allow for quality assessment of agents, improvement of agent skills and/or
dispute
resolution, for example.
[0002] In this regard, assessment of call quality is time consuming and very
subjective. For instance, a telephone call may last from a few seconds to a
few hours
and may be only one part of a customer transaction or may include several
independent transactions. The demeanor of the caller is also influenced by
events
preceding the actual conversation - for example, the original reason for the
call; the
time spent waiting for the call to be answered or the number of times the
customer has
had to call before getting through to the right person.
[0003] Assessing the "quality" of a telephone call is therefore difficult and
subject to
error, even when done by an experienced supervisor or full-time quality
assessor.
Typically, the assessment of a call is structured according to a pre-defined
set of criteria
and sub-criteria. Some of these may relate to the initial greeting, the
assessment of the
reason for the call, the handling of the core reason for the call, confirming
that the caller
is satisfied with the handling of the call, and leaving the call.
[0004] Automation of the assessment process by provision of standardized forms
and
evaluation profiles have made such assessment more efficient, but it is still
impractical
to assess more than a tiny percentage of calls. Moreover, even with a
structured
1
CA 02600535 2007-09-28
evaluation form, different assessors will evaluate a call differently with
quite a wide
variation of scores.
SUMMARY
[0005] In this regard, systems and methods for analyzing communication
sessions
using fragments are provided. An embodiment of a method comprises: delineating
fragments of an audio component of a communication session, each of the
fragments
being attributable to a party of the communication session and representing a
contiguous period of time during which that party was speaking; and
automatically
assessing quality of at least some of the fragments such that a quality
assessment of
the communication session is determined.
[0006] An embodiment of such a system comprises a communication analyzer
operative to: delineate fragments of an audio component of a communication
session,
each of the fragments being attributable to a party of the communication
session and
representing a contiguous period of time during which that party was speaking;
and
automatically assess quality of at least some of the fragments such that a
quality
assessment of the communication session is determined.
[0007] Computer readable media also are provided that have computer programs
stored thereon for performing computer executable methods. In this regard, an
embodiment of such a method comprises: delineating fragments of an audio
component of a communication session, each of the fragments being attributable
to a
party of the communication session and representing a contiguous period of
time
during which that party was speaking; and automatically assessing quality of
at least
some of the fragments such that a quality assessment of the communication
session is
determined.
2
CA 02600535 2007-09-28
[0008] Other systems, methods, features and/or advantages of this disclosure
will be
or become apparent to one with skill in the art upon examination of the
following
drawings and detailed description. It is intended that all such additional
systems,
methods, features, and advantages be included within this description and be
within
the scope of the present disclosure.
BRIEF DESCRIPTION
[0009] Many aspects of the disclosure can be better understood with reference
to the
following drawings. The components in the drawings are not necessarily to
scale,
emphasis instead being placed upon clearly illustrating the principles of the
present
disclosure. Moreover, in the drawings, like reference numerals designate
corresponding parts throughout the several views. While several embodiments
are
described in connection with these drawings, there is no intent to limit the
disclosure
to the embodiments disclosed herein.
[0010] FIG. 1 is a schematic diagram illustrating an embodiment of a system
for
analyzing communication sessions using fragments.
[0011] FIG. 2 is a flowchart illustrating functionality (or methods steps)
that can be
preformed by the embodiment of the system for analyzing communication sessions
using fragments of FIG. 1.
[0012] FIG. 3 is a schematic representation of an exemplary communication
session
and corresponding call fragments.
[0013] FIG. 4 is a flowchart illustrating functionality (or methods steps)
that can be
preformed by another embodiment of a system for analyzing communication
sessions
using fragments.
3
CA 02600535 2007-09-28
[0014] FIG. 5 is a diagram illustrating an embodiment of a system for
analyzing
communication sessions using fragments that is implemented by a computer.
DETAILED DESCRIPTION
[0015] Systems and methods for analyzing communication sessions using
fragments
are provided. In this regard, several exemplary embodiments will be described
in
which a recording of a telephone call is divided into more manageable
fragments. By
way of example, each of the fragments can be configured as contiguous speech
of a
party of the call. Specific behaviors can, therefore, be identified
automatically as each
fragment can be assessed more easily and unambiguously than if the behaviors
were
attempted to be identified from within an undivided call. By automating the
assessment
of call quality, a higher proportion of calls can be analyzed and hence a
higher
proportion of problem behaviors, processes and issues identified and addressed
with less
effort and cost than alternative manual strategies.
[0016] In this regard, FIG. 1 is a schematic diagram illustrating an
embodiment of a
system for analyzing communication sessions using fragments. As shown in FIG.
1,
system 100 incorporates a communication analyzer 110 that is configured to
analyze
audio components of communications. In FIG. 1, the audio component (not shown)
is
associated with a communication session that is occurring between a caller 112
and an
agent 114 via a communication network 116. In this embodiment, the agent is
associated with a contact center that comprises numerous agents for
interacting with
customers, e.g., caller 112.
[0017] One should note that network 116 can include one or more different
networks
and/or types of networks. As a non-limiting, example, communications network
116
can include a Wide Area Network (WAN), the Internet, and/or a Local Area
Network
4
CA 02600535 2007-09-28
(LAN). Additionally, the communication analyzer can receive information
corresponding to the communication session directly or from one or more
various
components that are not illustrated in FIG. 1. By way of example, the
information can
be provided from a long term storage device that stores recordings of the
communication session, with the recordings being provided to the storage
device by a
recorder. Additionally or alternatively, the recordings could be provided
directly from
such a recorder.
[0018] In operation, the analyzer of FIG. 1 performs various functions (or
method
steps) as depicted in the flowchart of FIG. 2. As shown in FIG. 2, the
functions
include (as depicted in block 210) delineating an audio component of a
communication session into fragments. In particular, in this embodiment, each
of the
fragments is attributable to a party of the communication session and
represents a
contiguous period of time during which that party was speaking. By way of
example,
one such fragment could involve a recording (e.g., 4 seconds in duration) of
the
speech of agent 114 during a communication session with customer 112, in which
the
agent greeted the customer. As shown in block 212, the analyzer also
automatically
assesses quality of at least some of the fragments such that a quality
assessment of the
communication session is determined.
[0019] In some embodiments, the parties to a communication session are
recorded
separately. In other embodiments, a session can be recorded in stereo, with
one
channel for the customer and one for the agent.
[0020] A vox detection analyzer of a communication analyzer can be used to
determine when each party is talking. Such an analyzer typically detects an
audio
level above a pre-determined threshold for a sustained period (the "vox turn-
on
time"). Absence of audio is then determined by the audio level being below a
pre-
CA 02600535 2007-09-28
determined level (which may be different from the first level) for a pre-
determined
time (which may be different from the previous "turn-on" time). By identifying
audio
presence on each of the two channels of recording of a call results in a time
series
through the call that identifies who, if anyone, is talking at any given time
in the series.
[0021] Once audio presence is determined, the call can be broken into
"fragments"
representing the period in which each party talks on the call. In this regard,
a fragment
can be delimited by one or more of the following:
i) the start or end of the call;
ii) the other party starting to speak and the previous party stopping
speaking;
iii) a "significant" pause - a period greater than a typical interval between
one party
finishing speaking and the other party beginning speaking. This interval may
be pre-
determined or determined by examining the actual intervals between the parties
speaking on this call. If the call involves more than a few alternations of
which
party is speaking, these alternations can typically be grouped. For instance,
one
group could be "normal turns of dialog" in which the intervals are on the
order of a
fraction of a second to one or two seconds and another group could be "delays"
in
which the dialog is hesitant or significantly delayed for some reason; and
iv) a "significant interruption" - a period during which both parties are
speaking
and which is longer than typical confirmatory feedback (e.g., "uh-huh") that
is
heard every few seconds in a normal interaction.
[0022] A schematic representation of an exemplary communication session and
corresponding call fragments is depicted in FIG. 3. As shown in FIG. 3, the
communication session is a sequence of audio components (depicted as blocks)
of an
interaction between an agent and a customer that takes place over a 30 second
time
period. In particular, the agent speaks for the first 4 seconds, followed by a
1 second
6
CA 02600535 2007-09-28
pause. The customer then speaks for 7 seconds followed by a 1 second pause.
Thereafter, the agent speaks for 7 seconds, the last 2 seconds of which the
customer
begins speaking, with the customer continuing to speak for another 2 seconds.
After
another 1 second pause, the agent speaks for 5 seconds after which, without
pause, the
customer speaks for 2 seconds and the communication session ends. Notably,
although not shown in this example, the reason for delimiting the fragment can
be
correlated with the fragment itself (e.g., alongside the fragment) resulting
in a
sequence of records.
[0023] Having broken a call into fragments, the system can analyze the
sequence and
duration of the fragments. By way of example, for each fragment, some
embodiments
can determine one or more of the following:
i) which party is speaking (customer or agent);
ii) which party spoke in the previous fragment;
iii) which party speaks in the next fragment;
iv) the delay between the previous fragment and this one;
v) the delay between this fragment and the next;
vi) a link to the previous fragment;
vii) a link to the next fragment;
viii) a transcript of the words and/or phonemes contained within the fragment -
determined by phonetic analysis using a phonetic analyzer and/or speech
recognition analysis using a speech recognition engine;
ix) a time sequence of the amplitude of the audio of the speaking party
throughout the fragment;
x) an estimate of periods of loud speech or shouting. This may be determined
by the fact that the audio level clipped as well as or instead of exceeded a
7
CA 02600535 2007-09-28
specified level or relative level compared to the call as a whole or the level
of audio from the other party;
xi) the time from the start of the call to the start of this fragment;
xii) the duration of this fragment; and
xiii) the time from this fragment to the end of the call.
In some embodiments, statistics of the call can be deduced from the individual
call fragment data. These may include one or more of:
i) number of call fragments;
ii) number of times the speaker changed;
iii) average duration of customer speaking;
iv) average duration of agent speaking;
v) percentage of total talk time that agent spoke;
vi) percentage of total talk time that customer spoke;
vii) percentage of total call time during which neither party spoke;
viii) percentage of time that both parties spoke;
ix) maximum duration of "interruptions" - defined for example, as periods of
greater than 1 second during which both parties talked; and
x) emotion indication - for example, pitch values and/or trends throughout the
call.
[0024] As mentioned above, a communication analyzer can automatically assess
quality of a communication session by assessing quality of at least some of
its
fragments. In order to accomplish quality assessment, various techniques can
be used.
By way of example, fragment training can be used, in which manual scoring is
applied
to one or more fragments and then the system applies comparable scoring to
fragments that are evaluated to be similar.
8
CA 02600535 2007-09-28
[0025] In this regard, in some embodiments, individual fragments or sequences
of two
or more successive fragments are presented to the user of the system,
typically with a
clear indication of which party is speaking and the delay between the two
fragments.
The user listens to some or all of the fragments and then indicates, such as
via a form
on a screen provided by a scoring analyzer, whether the fragments relate to a
good,
bad or "indifferent" interaction, for example. In many cases, the isolated
fragments
will not indicate a particularly good or bad experience but in a small
percentage of
cases such fragments can indicate a particularly good or bad experience. By
way of
example, a long delay between two successive fragments can be considered "bad"
but
in other cases, the words uttered, the tone or volume of the utterance may
indicate a
good or bad experience. This manual (human) assessment of the quality of the
fragment sequence can be stored and used to drive machine learning algorithms.
[0026] In some embodiments, in contrast to a scoring of good, bad or
indifferent, a
continuous scale (e.g., 0-10 rating) can be used. Additionally, multiple
criteria may be
presented, each of which the user can choose to provide feedback on, such as
"Customer empathy" and "Persuasiveness" for example. In many cases, any
particular
fragment or fragment pair will not be particularly good or bad but as long as
those
cases that are at one extreme or the other are identified, the system will
receive
valuable input.
[0027] In many cases, however, the fragments presented to the user may not
show
anything significant but may indicate that the previous or next fragments may
provide
more valuable input. Because of this, the user may be presented with controls
that
allow the user to play the previous and/or next fragment. Thus, the user can
provide
feedback on those fragments and/or move on to the next or previous fragment.
9
CA 02600535 2007-09-28
[0028] Where users assess whole calls, the overall quality assessment of the
call and
the individual criteria/sub-criteria may be noted. These are then applied to
either all
fragments or, where specific criteria are explicitly linked to particular
regions of the
call (e.g. "Quality of Greeting", "Confirmation of resolution"), to the
fragments of the
call according to a weighting function. In those embodiments that use
weighting, a
different weighting can be applied to each fragment according to the distance
of that
fragment from the start of the call, the end of the call, or from some other
known point
within the call. It should be noted that point from which the fragment is
measured for
weighting purposes can be identified by an event that occurred during the
call. The
fragment can be subsequently stored with a timestamp linking the fragment to
that
point, e.g., event, in the call.
[0029] As mentioned before, manual quality assessments can then be used by the
system for enabling automated scoring of other fragments that have not been
manually
scored. Additionally or alternatively, some embodiments can be provided with a
number of heuristics, such as predefined rules, that the system can use during
automated analysis by a scoring analyzer. In this regard, such rules can
involve one or
more of the following:
i) calls in which the customer to agent speech ratio is > 80/20 or less than
20/80 are scored as "bad";
ii) interruptions of >1 second are "bad";
iii) delays between fragments of >2 seconds are "bad"; and
iv) audio volumes above X are "bad".
[0030] The human input, e.g., predefined rules and/or examples of manually
assessed
calls/fragments, can be used as input for a variety of machine learning
techniques such
as neural nets and Bayesian filters expert systems, for example. By
identifying the
CA 02600535 2007-09-28
characteristics of the call fragments that lead to the assessments given, a
system
employing such a technique can learn to identify the relevant characteristics
that
differentiate "good" from "bad" calls.
[0031] An example of this approach is a Bayesian probability assessment of the
content of a call fragment. In such an approach, a transcript of a call may be
processed and the frequency of the occurrence of each word within the
customer's
speech is stored. The proportion of "good" fragments in which each word occurs
and
the proportion of "bad" fragments in which each word occurs is then noted.
These
probabilities can then be used to assess whether other fragments are likely to
be
"good" or "bad" based on the words within those and the likelihood of each of
the
words to be found in a "good" or "bad" fragment. From the many words within a
given fragment, those that provide the strongest discrimination of good versus
bad
fragment can be used and the remainder discarded. Of the N strongest
indicators, an
overall assessment can be made of good versus bad.
[0032] Typically, the other attributes of a fragment, such as those described
above, can
be used as potential indicators of the good/bad decision. These inputs may be
provided
to train a neural network or other machine learning system.
[0033] In some embodiments, feedback can be used to further enhance analysis.
Specifically, since a high proportion of fragment sequences do not indicate
particularly
good (or bad) experiences, it can be beneficial if a system presents to a user
those
fragments that is has identified as good or bad. By presenting these fragments
and
showing the assessment (good or bad) that the system has determined, the user
can be
enabled to confirm or correct the assessment. This input can then be fed back
into the
training algorithm either reinforcing the correct assessment or helping to
avoid repetition
of the mistake made.
11
CA 02600535 2007-09-28
[0034] In this regard, FIG. 4 is a flowchart depicting functionality of an
embodiment of
a system that incorporates the use of feedback. As shown in FIG. 4, the
functionality (or
method) may be construed as beginning at block 410, in which a communication
session
is recorded. In block 412, an audio component of the communication session is
delineated as a sequence of fragments. In block 414, inputs (such as manual
scoring
of a subset of the fragments and/or heuristics) are received for enabling
automated
scoring of at least some of the fragments. In block 416, the inputs are used
in
analyzing the fragments such that scores for at least some of the fragments
that were
not manually scored are produced. It should be noted that in some embodiments,
the
fragments that are manually evaluated may not be associated with the
communication
session that is being automatically scored.
[0035] In block 418, scores produced during automated analysis are presented
to a
user for review. By way of example, the scores can be presented to the user
via a
graphical user interface displayed on a display device. Then, in block 420,
inputs
from the user either confirming or correcting the scores are provided, with
these
inputs being used to update the analysis algorithm of the communication
analyzer.
[0036] FIG. 5 is a schematic diagram illustrating an embodiment of a
communication
analyzer that is implemented by a computer. Generally, in terms of hardware
architecture, voice analyzer 500 includes a processor 502, memory 504, and one
or
more input and/or output (UO) devices interface(s) 506 that are
communicatively
coupled via a local interface 508. The local interface 508 can include, for
example but
not limited to, one or more buses or other wired or wireless connections. The
local
interface may have additional elements, which are omitted for simplicity, such
as
controllers, buffers (caches), drivers, repeaters, and receivers to enable
communications.
12
CA 02600535 2007-09-28
[0037] Further, the local interface may include address, control, and/or data
connections to enable appropriate communications among the aforementioned
components. The processor may be a hardware device for executing software,
particularly software stored in memory.
[0038] The memory can include any one or combination of volatile memory
elements
(e.g., random access memory (RAM, such as DRAM, SRAM, SDRAM, etc.)) and
nonvolatile memory elements (e.g., ROM, hard drive, tape, CDROM, etc.).
Moreover, the memory may incorporate electronic, magnetic, optical, and/or
other
types of storage media. Note that the memory can have a distributed
architecture,
where various components are situated remote from one another, but can be
accessed
by the processor. Additionally, the memory includes an operating system 510,
as well
as instructions associated with a speech recognition engine 512, a phonetic
analyzer
514, a vox detection analyzer 516 and a scoring analyzer 518. Exemplary
embodiments of each of which are described above.
[0039] It should be noted that embodiments of one or more of the systems
described
herein could be used to perform an aspect of speech analytics (i.e., the
analysis of
recorded speech or real-time speech), which can be used to perform a variety
of
functions, such as automated call evaluation, call scoring, quality
monitoring, quality
assessment and compliance/adherence. By way of example, speech analytics can
be
used to compare a recorded interaction to a script (e.g., a script that the
agent was to
use during the interaction). In other words, speech analytics can be used to
measure
how well agents adhere to scripts, identify which agents are "good" sales
people and
which ones need additional training. As such, speech analytics can be used to
find
agents who do not adhere to scripts. Yet in another example, speech analytics
can
measure script effectiveness, identify which scripts are effective and which
are not,
13
CA 02600535 2007-09-28
and find, for example, the section of a script that displeases or upsets
customers (e.g.,
based on emotion detection). As another example, compliance with various
policies
can be determined. Such may be in the case of, for example, the collections
industry
where it is a highly regulated business and agents must abide by many rules.
The
speech analytics of the present disclosure may identify when agents are not
adhering
to their scripts and guidelines. This can potentially improve collection
effectiveness
and reduce corporate liability and risk.
[0040] In this regard, various types of recording components can be used to
facilitate
speech analytics. Specifically, such recording components can perform one or
more
various functions such as receiving, capturing, intercepting and tapping of
data. This
can involve the use of active and/or passive recording techniques, as well as
the
recording of voice and/or screen data.
[0041] It should be noted that speech analytics can be used in conjunction
with such
screen data (e.g., screen data captured from an agent's workstation/PC) for
evaluation,
scoring, analysis, adherence and compliance purposes, for example. Such
integrated
functionalities improve the effectiveness and efficiency of, for example,
quality
assurance programs. For example, the integrated function can help companies to
locate appropriate calls (and related screen interactions) for quality
monitoring and
evaluation. This type of "precision" monitoring improves the effectiveness and
productivity of quality assurance programs.
[0042] Another aspect that can be accomplished involves fraud detection. In
this regard,
various manners can be used to determine the identity of a particular speaker.
In some
embodiments, speech analytics can be used independently and/or in combination
with
other techniques for performing fraud detection. Specifically, some
embodiments can
involve identification of a speaker (e.g., a customer) and correlating this
identification
14
CA 02600535 2007-09-28
with other information to determine whether a fraudulent claim for example is
being
made. If such potential fraud is identified, some embodiments can provide an
alert.
For example, the speech analytics of the present disclosure may identify the
emotions
of callers. The identified emotions can be used in conjunction with
identifying
specific concepts to help companies spot either agents or callers/customers
who are
involved in fraudulent activities. Referring back to the collections example
outlined
above, by using emotion and concept detection, companies can identify which
customers are attempting to mislead collectors into believing that they are
going to
pay. The earlier the company is aware of a problem account, the more recourse
options they will have. Thus, the speech analytics of the present disclosure
can
function as an early warning system to reduce losses.
[0043] Additionally, included in this disclosure are embodiments of integrated
workforce
optimization platforms, as discussed in U.S. Application No. 11/359,356, filed
on
February 22, 2006, entitled "Systems and Methods for Workforce Optimization."
which is hereby incorporated by reference in its entirety. At least one
embodiment of
an integrated workforce optimization platform integrates: (1) Quality
Monitoring/Call
Recording - voice of the customer; the complete customer experience across
multimedia touch points; (2) Workforce Management - strategic forecasting and
scheduling that drives efficiency and adherence, aids in planning, and helps
facilitate
optimum staffing and service levels; (3) Performance Management - key
performance
indicators (KPls) and scorecards that analyze and help identify synergies,
opnortunities and improvement areas; (4) e-Learning - training, new
information and
protocol disseminated to staff, leveraging best practice customer interactions
and
delivering learning to support development; and/or (5) Analytics - deliver
insights
from customer interactions to drive business performance. By way of example,
the
CA 02600535 2007-09-28
integrated workforce optimization process and system can include planning and
establishing goals - from both an enterprise and center perspective - to
ensure
alignment and objectives that complement and support one another. Such
planning
may be complemented with forecasting and scheduling of the workforce to ensure
optimum service levels. Recording and measuring performance may also be
utilized,
leveraging quality monitoring/call recording to assess service quality and the
customer
experience.
[0044] One should note that the flowcharts included herein show the
architecture,
functionality, and/or operation of a possible implementation of software. In
this
regard, each block can be interpreted to represent a module, segment, or
portion of
code, which comprises one or more executable instructions for implementing the
specified logical function(s). It should also be noted that in some
alternative
implementations, the functions noted in the blocks may occur out of the order.
For
example, two blocks shown in succession may in fact be executed substantially
concurrently or the blocks may sometimes be executed in the reverse order,
depending
upon the functionality involved.
[0045] One should note that any of the programs listed herein, which can
include an
ordered listing of executable instructions for implementing logical functions
(such as
depicted in the flowcharts), can be embodied in any computer-readable medium
for
use by or in connection with an instruction execution system, apparatus, or
device,
such as a computer-based system, processor-containing system, or other system
that
can fetch the instructions from the instruction execution system, apparatus,
or device
and execute the instructions. In the context of this document, a "computer-
readable
medium" can be any means that can contain, store, communicate, propagate, or
transport the program for use by or in connection with the instruction
execution
16
CA 02600535 2007-09-28
system, apparatus, or device. The computer readable medium can be, for example
but
not limited to, an electronic, magnetic, optical, electromagnetic, infrared,
or
semiconductor system, apparatus, or device. More specific examples (a
nonexhaustive list) of the computer-readable medium could include an
electrical
connection (electronic) having one or more wires, a portable computer diskette
(magnetic), a random access memory (RAM) (electronic), a read-only memory
(ROM) (electronic), an erasable programmable read-only memory (EPROM or Flash
memory) (electronic), an optical fiber (optical), and a portable compact disc
read-only
memory (CDROM) (optical). In addition, the scope of the certain embodiments of
this disclosure can include embodying the functionality described in logic
embodied
in hardware or software-configured mediums.
[0046] It should be emphasized that the above-described embodiments are merely
possible examples of implementations. Many variations and modifications may be
made to the above-described embodiments. All such modifications and variations
are
intended to be included herein within the scope of this disclosure.
17
